https://github.com/JayParanjape/LoRASAM

Code for Low-Rank Adaptation of SAM for Surgical Scene Segmentation
https://github.com/JayParanjape/LoRASAM

Last synced: 30 days ago
JSON representation

Code for Low-Rank Adaptation of SAM for Surgical Scene Segmentation

• Host: GitHub
• URL: https://github.com/JayParanjape/LoRASAM
• Owner: JayParanjape
• Created: 2023-12-22T12:48:48.000Z (about 1 year ago)
• Default Branch: master
• Last Pushed: 2023-12-22T12:50:18.000Z (about 1 year ago)
• Last Synced: 2023-12-22T14:55:49.446Z (about 1 year ago)
• Language: Python
• Size: 1.4 MB
• Stars: 0
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Segment-Anything - [code

README

        
# LoRASAM

This repository contains the code for **Low-Rank Adaptation of Segment Anything Model for Surgical Scene Segmentation**

## Environment File

Create a new conda environment with the config file given in the repository as follows:

```

conda env create -f lorasam.yml

conda activate lorasam

```

## General file descriptions

- data_transforms/*.py - data transforms defined here for different datasets.

- data_utils.py - functions to generate dataloaders for different datasets

- model.py - model architectures defined here

- prompt_adapted_segment_anything/modeling/lora_layers.py - code for the Low Rank modifications used in the model

- train.py - code for general training, common to all datasets

- driver_scratchpad.py - driver code for training models. 

- eval/*/generate_predictions.py - code for generating results for a given dataset

- eval/*/generate_predictions.sh - script to run generate_predictions for generating results for all labels of interest.

- model_loratuning.yml - config file for defining various model hyperparameters for LoRASAM

- config_.yml - config file for defining various dataset related hyperparameters

  

## Example Usage for Training

```

python driver_scratchpad.py --model_config model_loratuning.yml --data_config config_endovis.yml --save_path "./temp.pth"

```

Please refer to driver_scratchpad.py for other command line options and parameters.

## Example Usage for Evaluation

```

cd eval/endovis

bash generate_all_predictions.sh

```

## Citation

```

To be added

```

Please feel free to reach out to me or raise an issue in case of trouble while running the code.